Models are given yare's documentation and tasked to one-shot a winning bot. Each model gets 3 bots. Matchups are Pokemon-style: the winner stays, the loser swaps in their next bot. First to lose 3 bots loses the matchup.
Clowder Bot is a human-coded, yet relatively weak bot in the context of the game. All LLM models are generally doing poorly on this task for now.